Re: fillfactor and cluster table vs ZFS copy-on-write

On Thu, Apr 16, 2015 at 4:56 PM, Qingqing Zhou <zhouqq(dot)postgres(at)gmail(dot)com>
wrote:

> On Thu, Apr 16, 2015 at 5:09 AM, Geoff Speicher <gspeicher(at)umtechx(dot)com>
> wrote:
> > Therefore one might posit that PostgreSQL should be configured to use
> 100%
> > fillfactor and avoid clustering on ZFS. Can anyone comment on this?
> >
>
> Even with COW, I can see fillfactor < 100% still have its virtues. For
> example, HOT update can avoid adding an extra index item on the index
> page if it finds the new item can be inserted in the same heap page.
>

That's true, the new physical location on disk is transparent to the DBMS
so it has no more or less housekeeping with or without COW, but the
housekeeping still has to be done somewhere, so it helps to understand
which is more efficient. I'll see if I can produce some empirical data
unless anyone thinks it's a waste of time.

When you do CLUSTER command, engine will overwrite table into new
> files any way, so COW does not affect here.
>

I was thinking about CLUSTER more in terms of its positive side-effects to
the effective fillfactor but your point is taken.

Thanks,
Geoff